Professional Documents
Culture Documents
Final report
Approved for public release; distribution is unlimited
Berger, Rutherford C.
A finite element scheme for shock capturing / by R.C. Berger, Jr., ; prepared
for Assistant Secretary of the Army (R&D).
61 p. : ill. ; 28 cm. - (Technical report ; HL-93-12)
Includes bibliographical references.
1. Hydraulic jump - Mathematical models. 2, Hydrodynamics. 3. Shock
(Mechanics) - Mathematical models. 4. Finite element method. I. United
States. Assistant Secretary of the Army (Research, Development and Acquisi-
tion) 11. U.S. Army Engineer Waterways Experiment Station. Ill. In-house Labo-
ratory Independent Research Program (U.S. Army Engineer Waterways
Experiment Station) IV. Title. V. Series: Technical report (U.S. Army Engineer
Waterways Experiment Station) ; HL-93-12.
TA7 W34 no.HL-93-12
Contents
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1
Basic Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2
Shock equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 .
Shock relations in 2-D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9
2-Numerical . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .13
Approach
Advective Dominated Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
The Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Petrov-Galerkin formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . .14
Shock Capturing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .20
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .55
Preface
This report is the product of research conducted from January 1992 through
April 1993 in the Estuaries Division (ED), Hydraulics Laboratory (HL), U.S.
Army Engineer Waterways Experiment Station (WES), under the In-House
Laboratory Independent Research (ILIR) Program. The funding was providing
by ILIR work unit "Finite Element Scheme for Shock Capturing."
Dr. R. C. Berger, Jr., ED, performed the work and prepared this report
under the general supervision of Messrs. F. A. Herrmann, Jr., Director, HL;
R. A. Sager, Assistant Director, HL; and W. H. McAnally, Chief, ED.
At the time of publication of this report, Director of WES was Dr. Robert
W. Whalin. Commander was COL Bruce K. Howard, EN.
ntroduction
Background
Shocks in fluids result from fluid flow that is more rapid than the speed of
a compression wave. Thus there is no means for the flow to adjust gradually.
Pressure, velocity, and temperatures change abruptly, causing severe fatigue
and component destruction in military aircraft and engine turbines. This
problem is not limited to supersonic aircraft; many parts of subsonic craft are
supersonic. For example, the rotors of helicopters have supersonic regions as
do the blades of the turbine engines used on many crafts. The shock is formed
as the flow passes from supersonic to subsonic or, in the case of an oblique
shock, as the result of a geometric transition in supersonic flow. Wind tunnels
are limited in the Mach numbers they can achieve and testing is expensive;
thus design relies upon numerical modeling. In Gdraulics the equivalent
shocks are referred to as hydraulic jumps, surges, and bores. Here, for
example, it is important to determine the ultimate height of water resulting
from a dam break or the insertion of a bridge in a fast-flowing river.
The compressible Euler equations describe these flow fields and are solved
numerically in discrete models. These partial differential equations implicitly
assume a certain degree of smoothness in the solution. Models, therefore,
have great difficulty handling shocks. One method to avoid solving
numerically across the shock is to track the shock and impose an internal
boundary there. This method is called "shock-tracking." On the other hand
the sharp resolution of the shock can be forfeited and allow for O(1) error at
the transition. This is referred to as "shock-capturing," as originated by
von Neumann and Richtmyer (1950), and is now the most common technique
used in engineering practice.
Great care must be undertaken to make sure the errors are local to the
shock. Otherwise the shock location and speed will be incorrect. It is
important that the discrete numerical operations preserve the Rankine-Hugoniot
condition (Anderson, Tannehill, and Plekher 1984) resulting from integration
by parts. While this should result in reasonable shock speed and location,
discrete models commonly suffer from numerical oscillations near the shock.
There are many proposed methods used to reduce these oscillations. The
Chapter 1 Introduction
basic theme is to cleverly apply artificial diffusion as a result of flow
parameters. Many of these methods do not preserve the original equations
within the shock due to this added diffusion. Hughes and Brooks (1982) have
approached this problem within the finite elements method by the development
of a single test function that reflects the speed of fluid transport (the SUPG
scheme, Streamline llpwind Eetrov-Galerkin) to be applied to the entire partial
differential equation set. In this manner the model is consistent even at
discrete scales. Its application, thus far, has been only to the very simple case
of Burgers' equations.
In this report a two-dimensional (2-D) finite element model for the shallow-
water equations is produced using an extension of the SUPG concept, but rely-
ing upon the characteristics of the advection matrix (transport as well as the
free-surface wave speeds) to develop the test function to be applied to the
coupled set of equations. The shallow-water equations are a direct analogy to
the Euler equations with the depth substituted for density and dropping the
energy equation. This equation set is ideal for testing numerical schemes for
the Euler equations. The model developed can reproduce supercritical and
subcritical flow and is shown to reproduce very difficult conditions of
supercritical channel transitions and preserve the Rankine-Hugoniot conditions.
For simple geometries, analytic and flume results are compared with
approaches for shock-capturing and shock detection. A trigger mechanism that
turns on the capture schemes in the vicinity of shocks and the characteristic
upstream weighted test function are tested.
Basic Equations
The basic equations that are addressed are the 2-D shallow-water equations
given as:
Chapter 1 Introduction
where
where
t = time
h = depth
Chapter 1 Introduction
au
a,, = 2pv -
ax
p = fluid density
z = bed elevation
n = Manning's coefficient
These equations neglect the Coriolis effect and assume the pressure distribution
is hydrostatic, and the bed slope is assumed to be geometrically mild though it
may be hydraulically steep. These equations apply throughout the domain in
which the solution is sufficiently smooth. Now consider the case for which
the solution is not smooth.
Shock equations
In this section we first derive the jump conditions in one dimension (1-D)
with no dissipative terms, i.e., friction or viscosity. We show that as a result
of the discontinuity of the jump, the shallow-water equations should conserve
mass and momentum balance but will lose energy. Furthermore, if there is no
discontinuity, energy too will be conserved. Later the jump relations are
extended to 2-D.
This derivation relies upon the work of Stoker (1957) and Keulegan (1950)
following a fluid element through a moving jump. Figure 1 defines these
features.
If our coordinate system is chosen to move with the jump at speed V,, then
we may use the following term definitions.
Chapter 1 Introduction
Figure 1. Definition of terms for a moving hydraulic jump
Uo = velocity at section 0
ho = depth at section 0
U1 = velocity at section 1
hl = depth at section 1
p(Vlhl - Vehe) = 0
Chapter 1 Introduction
where q is the relative discharge. Equation 3 may be written in a fixed grid as
where the symbol [ I implies the jump in the quantities across the discon-
tinuity, e.g., b] H hl - ho.
where
V , [Uh] = [u2h + F]
If we multiply (8) by V , add this to (10) using the relationship for q we have:
Chapter 1 Introduction
Now substituting Equation 8 yields
or, finally
so that the shallow-water equations lose energy at the jump, and it is propor-
tional to the depth differential cubed. If the depth is continuous, no energv
will be lost.
Chapter 1 Introduction
This is completely arbitrary, and this
+ form also produces an easy test case
++ that will be eventually applied to the
numerical model. With U1 = 0,
If we consider the speed of the perturbation in front of and behind the shock,
we note that both move toward the shock.
To demonstrate this, we calculate the relative speeds Vo and V1. These are
the speeds of fluid particles as perceived by an observer moving with the
shock. We have already shown that Vl = -Vw or
If this value is negative, then a perturbation behind the jump catches the shock,
and from Equation 17 we know dgh, is greater in magnitude than V1, Wl c 0.
Chapter 1 Introduction
Again we calculate the relative speed of a perturbation, but now in front of
the shock:
Now if Wo is positive the shock catches up with the wave perturbation, and
since Vo is clearly greater than dgh, this is indeed what happens. Therefore
any small perturbations are swept toward, and are engulfed in the shock.
Chapter 1 Introduction
which may be expanded as
where,
where,
Chapter 1 Introduction
where
which states that the relative mass flux jump across the shock in the direction
n should be zero.
Chapter 1 Introduction
which for an arbitrary length T, to preserve the equation, the integrand itself
must satisfy the equation, therefore:
where,
Q' = V-h-
Q+ = v+ht
or
which states that the relative momentum flux in the direction n is balanced by
the pressure jump across the shock.
Chapter 1 Introduction
2 Numerica Approach
The quality of the numerical solution depends upon the choice of the basis
(or interpolation) function and upon the test function. The basis function
determines how the variable (or solution) is represented and the test function
determines the way in which the differential equation is enforced. Finite ele-
ments are a subset of the weighted residual method. Here one looks at the
solution of a differential equation in a weighted average sense. In the Galerkin
approach the test function is identical to the basis function. This method can
have difficulty with advection-dominated flow. The basic problem is that the
form of the test function (typically an even or symmetric function) cannot
detect the presence of a node-to-node oscillation, since this "spurious solution"
has a spatial derivative which is an odd function (antisymmetric). One
approach to resolve this problem is to use a mixed interpolation where, for the
shallow-water equations, the depth uses a lower order basis than does the
velocity (see, e.g., Platzman (1978) or Walters and Carey (1983)). Typically,
these are chosen as depth as an elemental constant and velocity as linear, or
depth linear and velocity as a quadratic. This approach effectively decouples
the depth from this node-to-node oscillation but depends upon some additional
artificial viscosity to damp velocity oscillations if the flow is not highly
resolved. Another approach is to modify the test function so that it includes
odd functions as well as even functions so that these modes can be detected
and if weighted properly, eliminated. Any approach in which the test function
differs from the basis function is termed a Petrov-Galerkin approach. In our
case we choose the Lagrange basis functions to be CO;i.e., the functions are
continuous. Let us consider an example to illustrate the problem with the
Calerkin approach and an approach to develop a Petrov-Galerkin test function.
First we will illustrate the problem that discrete formulations have with
advection-dominated flow. In this regard the 1-D linearized inviscid Burgers'
equation may be written
where the subscripts t and x represent partial derivatives with respect to time
and space, respectively, and
here c(x) is the approximate solution, and the subscript j indicates nodal
values and @j is the Gaierkin test function at node j.
Our numerical solution equation, for the steady-state problem (Ct=O) may
be written as the inner product
where
Cf(x), d-4) = SL f
0 g(x) dx
and the prime indicates the derivative with respect to x.
(Note that finite difference methods using central differences give an identical
result .)
Now, if the test function includes both odd and even components, this
mode will no longer be a solution. In fact, if we weight the test function
upstream, these oscillations are damped; weighting downstream amplifies them.
A common approach is to use a test function, q,weighted as follows,
Here the spatial derivative supplies the odd component to the test function.
The resulting discrete solution using this test function is
If a r 112 we will have no negative roots and therefore should not have a
node-to-node oscillation. This spurious root that we damp by increasing the
coefficient a is driven by some abrupt change, most notably when some dis-
continuity is required in the equations due to the imposition of boundary
conditions. It is more precise in a smooth region for smaller a.
where
where A is the matrix of eigenvalues of A, and P and P-' are composed of the
eigenvectors.
The size and direction of the added odd function is then based upon the
magnitude and direction of the characteristics.
where
where
and
A1=U+C
h;?=u-C
A3 = U
C = (gh)1t2
Shock Capturing
In the section, "Shock equations," in Chapter 1 we have shown that unless
there is a discontinuity in depth, mechanical energy will be conserved in the
shallow-water equations (with no friction or diffusion). So the obvious ques-
tion is what happens in a numerical scheme in which the depth is approxi-
mated as CO;i.e., it is continuous. We are onIy enforcing mass and
momentum, but we are implicitly enforcing energy conservation. This is the
result that the Galerkin approach will give using COdepth approximation. The
result is that while mass and momentum conservation are enforced over our
discrete model, energy is also conserved by including the spurious node-to-
node mode we discussed. Since energy involves v2 terms and momentum
only k: both can be satisfied in a weighted average sense over the region
included in the test function. This is due to
We have shown that when a shock is encountered, the weak solution of the
shallow-water equations must lose mechanical energy. Some of this energy
loss is analogous to a physical hydraulic system losing energy to heat, particle
rotation, deformation of the bed, etc; but much of it is, in fact, simply the
energy being transferred into vertical motion. And since vertical motion is not
included in the shallow-water equations, it is lost. This apparent energy loss
can be used to our advantage.
We would like to apply a high value of a, say 0.5, only in regions in which
it is needed, since a lower value is more precise. Therefore, we wish to con-
struct a trigger mechanism which can detect shocks and increase a automati-
cally. The method we employ detects energy variation for each element and
flags those elements which have a high variation as needing a larger value of
a for shock capturing. Note that this would work even in a Galerkin scheme
since this trigger is concerned with energy variation on an element basis and
the Galerkin method would enforce energy conservation over a test function
(which includes several elements).
where
ED; - E
Tsi =
S
SZi = element i
E = mechanical energy
a; = area of element i
and
E = the average element energy over the entire grid
The testing of this scheme and model behavior was undertaken in stages.
These progress from what is essentially a 1-D test for shock speed which can
be determined analytically, to a 2-D dam break type problem comparison with
flume data, to more general 2-D geometry comparison of supercritical transi-
tion in a flume but for steady state. This series tests the model against the
analytic results of the shallow-water equations for very limited geometry, and
progresses to more general geometry with the limitation of the shallow-water
equations in reproducing actual flow problems. The applicability of the
shallow-water equations to these flume conditions is not so important in this
study (since it is interested in shock capturing), but is important for model
application in open-channel hydraulics.
The second case is a comparison to a flume data set reported in Bell, Elliot,
and Ghaudhry (1992) which is analogous to a dam break problem. Here the
shock is in a horseshoe-shaped channel and the comparison is to actual flume
data. The comparisons are made to the water surface heights and timing of the
shock passage.
Chapter 3 Testing
Case 1: Analytic Shock Speed
The shock speed for the shallow-water equations given simple 1-B
geometry can be determined analytically. These are the Rankine-Nugoniot
relations shown in Equations 5 and 9. This provides a direct comparison with
the model shock speed without relying upon hydraulic flume data, for which
discrepancy will be due to the hydrostatic assumption made in the shallow-
water equations. Instead we have a direct way of evaluating the numerical
scheme alone. As spatial and temporal resolution increase, the numerical
shock speed should converge to the analytic speed. The test consists of setting
a supercritical flow in a long channel, closing the downstream end, and
calculating the speed of the jump that forms and propagates upstream. The
initial conditions for this test case are shown in Table 1. The test conditions
are shown in Table 2. The term at indicates the
method applied to the temporal derivative, 1.0 is first-order backward, and 1.5
is second-order backward. The subscript s indicates the value in the shock
vicinity. The a and a, are the weighting of the Petrov-Galerkin contribution
throughout the domain and in the shock vicinity, respectively. With
Manning's n and viscosity of 0.0 there is no dissipation in the shallow-water
equations.
Figures 6-8 and 9-11 show the center-line profile over time of these tests
for at = 1.0 and for AX = 0.4 and 0.8 m, respectively. These plots represent
the center-line depth profile over time in a perspective view. The vertical axis
is the flow depth, the horizontal axis is time, and the axis that appears to be
into the page is the distance along the channel. From this one can see that as
Chapter 3 Testing
Figure 6. Time-history of center-line water surface elevation profiles; 9 = 1.0, Ax = 0.4 m, At =
0.4 sec
Chapter 3 Testing
Figure 8. Time-history of center-line water surface elevation profiles; 9 = 1.0, Ax = 0.4 m, At =
1.6 sec
Chapter 3 Testing
Figure 10. Time-history of center-line water surface elevation profiles; at = 1.O, Ax = 0.8 m, At
= 1.6 sec
Figure 11. Time-history of center-line water surface elevation profiles; o+ = 1.O, Ax = 0.8 m, At
= 3.2 sec
Chapter 3 Testing
one moves over time, the center-line profile shock moves upstream. It is apparent that as the
spatial and temporal resolution improve, the shock becomes steeper. The shock is fairly
consistently spread over three or four elements; and so as the element size is reduced, the
resulting shock is steeper. The x-t slope of the shock indicates the shock speed. Any bending
would indicate that the speed changed over time, which should not be the case. The upper
elevation is precisely 0.2 m, which is correct. There is no overshoot of the jump, though there
is some undershoot when C, is less than 1. Cs is the product of the analytic shock speed and
the ratio of time-step length to element length. A C, value of 1 indicates that the shock
should move 1 element length in 1 time-step.
Figures 12 and 13 show the error in calculated speed and the relative error in calculated
speed, respectively. These are for AX = 0.4, 0.8 and 1.0 m which is reflected in the Grid
Resolution Number defined as MlAh. Here h is the depth and Ah is the analytic depth
difference across the shock, 0.1 m. The error was a s small as was detectable by the technique
for measurement of speed at AX = 0.4 m so there was no need to go to smaller grid spacing.
Values of C, less than 1 appear to lag the analytic shock and Cs greater than 1 leads the
analytic shock. With the largest C, the calculated shock speed is greater than the analytic by
at most 0.0034 mlsec which is only 0.6 percent too fast. As resolution is improved the
solution appears to converge to the analytic speed.
Figures 14-16 and 17-19 are the center-line profile histories for at = 1.5 and for AX = 0.4
and 0.8 m, respectively. It is apparent that the lower dissipation from this second-order
scheme allows an oscillation which is most notable upstream of the jump for larger values of
C,. But as C, decreases, there is an undershoot in front of the shock. The slope of the x-t
line along the top of the shock has a significant bend early in the high Cs simulations. The
speed is too slow here.
Now consider the associated Figures 20 and 21 for error in calculated shock speed and
relative error in calculated speed. The error is actually worse than for the first-order scheme.
This is due primarily to the slow speed early in the simulation; if this is dropped by using only
the last 50 seconds of simulation, the relative error is only 0.6 percent slower than analytic.
Once again, as the resolution improves, the solution converges to the proper solution.
Chapter 3 Testing
Model Shock Speed Precision
0.01
Cs = 2.191
2 Cl
g 0
A 8%
0
0 Cs = 1.095
W
0 Cs = 0.548
-0.01 I I I I I
0 CI d \O
" 2 2
Figure 12. Error in model shock speed with grid refinement for at = 1.0
0.02 +
B
a
V)
3
n
V)
13
-
.-*
A
0
Cs = 2.191
.
4
8
0
0 -
A A
0
0 CS = 1.095
t:
W 0 Cs = 0.548
'a
R.
V)
3n
V)
-0.02 I 1 I I I
0 CI d '0 " S 2
Figure 13. Relative error in model shock speed with grid refinement for at =
1.o
Chapter 3 Testing
Figure 14. Time-history of center-line water surface elevation profiles; 9 = 1.5, Ax = 0.4 m, At
= 0.4 sec
Figure 15. Time-history of center-line water surface elevation profiles; 9 = 1.5, Ax = 0.4 m, At
= 0.8 sec
Chapter 3 Testing
Figure 16. Time-history of center-line water surface elevation profiles; 9= 1.5, Ax = 0.4 m, At
= 1.6 sec
Figure 17. Time-history of center-line water surface elevation profiles; 3 = 1.5, & = 0.8 m, At
= 0.8 sec
Chapter 3 Testing
31
Figure 18. Time-history of center-line water surface elevation profiles; 3 = 1.5, Ax = 0.8 m, At
= 1.6 sec
Figure 19. Time-history of center-line water surface elevation profiles; 9 = 1.5, Ax = 0.8 m, ~t
= 3.2 sec
Chapter 3 Testing
Model Shock Speed Precision
0.01
0 Cs = 2.191
2 0 0
-
g
W
0 w
A
0 V 0 Cs = 1.095
0 Cs = 0.548
-0.01 I I I I I
0 2 4 6 8 10 12
Figure 20. Error in model shock speed with grid refinement for 9 = 1.5
Figure 21. Relative error in model shock speed with grid refinement for at =
1.5
Chapter 3 Testing
Chapter 3 Testing
The numerical grid is shown in Figure 23, and contains 698 elements and
811 nodes. This grid was reached by increasing the resolution until the results
no longer changed. The most critical reach is in the region of the contraction
near the dam breach. The basic element length in the channel is 0.1 m and
there are five elements across the channel width. For the smooth channel case,
Bell, Elliot, and Chaudhry (1992) used a 1-D calculation to estimate the
Manning's n to be 0.016 but experience at the Waterways Experiment Station
suggests that this value should actually be 0.009, which seems more
reasonable.
The test results for stations 4, 6 and 8 are shown in Figures 24-26. Here
the time-history of the water elevation is shown for the inside and outside of
the channel for both the numerical model (at 5 of 1.0 and 1.5) and the flume.
The inside wall is designated by squares and the outside by diamonds. Of
particular importance is the arrival time of the shock front. At station 4 the
numerical prediction of arrival time using 5 of 1.0 is about 3.4 sec which
appears to be about 0.05 sec sooner than for the flume. This is roughly
1-2 percent fast. For 9 of 1.5 the time of arrival is 3.55 sec which is about
0.1 sec late (3 percent). At station 6 both flume and numerical model arrival
times for at of 1.0 were about 4.3 sec and for slation 8 the numerical model is
5.6 sec and the flume is 5.65 to 5.8 sec. With % set at 1.5 the time of arrival
is late by about 0.2 and 0.15 sec at stations 6 and 8, respectively. The flume
at stations 6 and 8 has a earlier arrival time for the outer wave connpared to
the inner wave. The numerical model does not show this. In comparing the
water ellevations between the flume and the numerical model, it is apparent that
the flume results show a more rapid rise. The numerical model is smeared
somewhat in time, likely as a result of the first-order temporal derivative
calculation of 5 of 1.0. The numerical model with at set at 1.5 shows the
overshoot that was demonstrated in Case 1. This is likely a numerical artifact
and not based upon physics even though this looks much like the flume
results. The surge elevations predicted by the numerical modd are fairly close
if one notices that the initial elevation of the flume data is supposed to be
0.0762 m and it appears to be recorded as much as 0.015 rn higher at some
gages. Since the velocity is initially zero then all of these readings should
have been 0.0762 m and all should be adjusted to match this initial elevation.
Chapter 3 Testing
Chapter 3 Testing
Time, sec
e e ~ . . 0 0 ~ . . 4 0 ~ 4 b 4 ~ e e b ~ ~ e . e ~ . o ~ e e e o ~ ~
T b c , sec
.e...4.**..*.*4*.4*.......,....4e4<
'a *0~~~~000000~000~0000~0000000000000011
3 0.15- (I
8 0
Inner wave
.
o
Time, sec
Figure 24. Flume and numerical model depth histories for station 4
Chapter 3 Testing
Figure 25. Flume and numerical model depth histories for station 6
Chapter 3 Testing
Station 8, Flume
T i e , sec
T i e , sec
T i e , sec
Figure 26. Flume and numerical model depth histories for station 8
Chapter 3 Testing
With this in mind, stations 4 and 8 match fairly closely between flume and
numerical model. Station 4 in the flume would still have a greater difference
between outer and inner wave than that predicted by the model. The differ-
ence might be a manifestation of a three-dimensional effect that the model
cannot mimic. The overall timing and height comparisons are good.
Figure 27 shows the spatial profile of the outer wall water surface elevation
of the numerical model versus distance downstream from the dam. These
distance measurements are in terms of the center-line distance. The two condi-
tions are for cq of 1.0 and 1.5, i.e., first- and second-order temporal derivative.
The nodes are delineated by the symbols along the lines. The overshoot of
the second-order scheme and the damping of the first-order is obvious. Again,
it is probable that the overshoot is a numerical artifact even though this is
much like what the flume would show.
Chapter 3 Testing
(0.3048 m) wide in a length of 4.78 ft (1.457 m), i.e., an angle of 6 deg on
each side.
The model resolution was increased until we were confident that the results
no longer changed with greater resolution. The numerical model was set up
with 10 evenly spaced elements laterally across the channel and 24 over the
length of the transition. The model limits were extended some 40 ft
(12.192 m). The total number of nodes was 1661 with 1500 elements. As in
the flume test the numerical model was set up to provide a uniform depth of
0.1 ft (0.03048 m) approaching the transition. The bed slope chosen was
0.05664. The other parameters are shown in Table 4.
The results from the numerical model run and the flume results are shown
in Figure 28. The oblique shock forms along the sidewalls of the transition
and impinges on the point in which the converging channel goes back to paral-
lel walls. This, by the way, is the manner in which one would want to design
a lateral transition. The positive wave from the beginning of the converging
walls will tend to cancel the negative wave originating at the point where the
walls change back to parallel. The heights of the water surface are indicated
by the contours in both model and flume. The maximum and minimum
heights compare fairly well. The shape is good as well. Generally the wave
from the shallow-water equation will be swept downstream less than that from
the flume results since the shallow-water equations will transport all wave-
lengths at the speed of a long wave. Shorter waves will travel more slowly
than the shallow-water equations predict. The comparison is good, and the
model demonstrates that the shock capturing technique functions well in a
general 2-D setting.
Chapter 3 Testing
0.5 0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0
DISTANCE FROM CONTRACTION, FT
0.5 0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0
DISTANCE FROM CONTRACTION, FT
Figure 28. Comparison of flume and numerical model water surface elevations for the super-
critical transition case, straight-wall contraction F = 4.0. To convert feet to
meters, multiply by 0.3048
Chapter 3 Testing
Discussion
Now let's study the behavior of the 1-D linearized shallow-water equation
analytically and numerically. This could lead to a conceptual appreciation of
the behavior we have observed in the testing section of the report. We shall
follow a Fourier analysis of the wave components; for examples, see
Leendertse (1967) or Froehlich (1985). First let's consider the nondimension-
alized shallow-water equations
such that P : A ~ P , = A,
where
Chapter 3 Testing
A, is the diagonal matrix of eigenvalues and Po and P-: are composed of the
eigenvectors (and are arbitrary). With the substitution of Equation 55 into 54
and multiplication by P-: we retrieve the diagonalized shallow-water equations
in terms of the Riemann Invariants
A
where T is a constant and K is the wave number, we arrive at the solution
where
o = m, the wave frequency
y = -io
With this solution we shall now compare the behavior of the model to that of
the analytic solution.
and
Chapter 3 Testing
where the superscript n indicates the time-step and the subscript j is the spatial
node location.
We now present the results of this analysis for a = 112 and for the temporal
derivative parameter at of 1.0 and 1.5. We shall compare the relative ampli-
tude and relative speed for a single time-step. The parameter for relative speed
is given by
tan
relative speed =
where
AAt,
C = Courant number r -
Ax*
The amplitude portrait shows substantial damping for larger C and for the
shorter wavelengths (or alternatively the poorer resolution). The large damp-
ing at a wavelength of 2Ax is important, as this is the mechanism that provides
the energy dissipation to capture shocks. Now consider the phase portrait, or
in this case the relative speed portrait. Over the conditions shown, the numeri-
cal speed is less than the analytic speed throughout. For larger C the relative
speed is somewhat lower (worse). For N = 2 the speed is 0, so that undamped
oscillation could remain at steady state.
Chapter 3 Testing
Figure 29. Relative amplitude versus C and resolution for at = 1.0 and a = 0.5
Figure 30. Relative speed versus C and resolution for at = 1.0 and a = 0.5
Chapter 3 Testing
In comparison to the results we have shown in Figures 6-11 for Case 1,
analytic shock case, we must remember that C, is the Courant number based
on shock speed, whereas C is based on the perturbation wave speed. If we
consider a wave moving upstream just behind the shock, since short wave-
lengths move t ~ slowly,
o the disturbance of the shock produces waves of these
length which fall behind the shock rather than remaining within. As the time-
step is reduced (C, gets smaller) the relative speed is better for the moderate
wavelengths and so the shock front becomes sharper.
At a point near the shock front we note that generally w e get a sharp front
with no undershoot until we reach the smallest time-step. Again if we are
within the shock at a depth where there is an upstream propagating wave
(subcritical), is there a Courant number C that has a relative speed greater than
analytic. This would be the only way in which an undershoot could appear.
Figure 31 extends the relative wave speed portrait below C = 0.5. From this
figure it is apparent for small values of C that the numerical wave speed is
greater than analytic so that it is possible to develop an undershoot in front of
the jump.
For the sake of completeness the relative amplitude and speed portraits are
included for a = 0 and 0.25 at at of 1.0 and 1.5 in Figures 35-42. The condi-
tion a = 0 is, in fact, the Galerkin case since the Petrov-Galerkin contribution
is included through a. The Galerkin approach is shown to contain a steady-
state spurious mode due to the speed of zero for N = 2. Furthermore, this
mode is undamped. The case of a = 0.25 shows that the relative speed
portraits change very little from a = 0.5 but the amplitude damping is
improved.
The obvious conclusions that can be drawn from this discussion is that for
an unsteady run either use at = 1.5 or take smaller time-steps with at = 1.0.
An improvement in spatial resolution dramatically improves the solution.
Chapter 3 Testing
Relative Speed 0 -
Figure 31. Relative speed versus C and resolution for at = 1.0 and a = 0.5, for
small values of C
Relative Amplitude 0.
Figure 32. Relative amplitude versus C and resolution for at = 1.5 and a = 0.5
Chapter 3 Testing
Relative Speed 0 .
Figure 33. Relative speed versus C and resolution for at = 1.5 and a = 0.5
Relative Speed 0 .
10
Elements per Wavelength
Figure 34. Relative speed versus C and resolution for at = 1.5 and a = 0.5, for
small values of C
Chapter 3 Testing
Figure 35. Relative amplitude versus C and resolution for at = 1.0 and a = 0
Figure 36. Relative speed versus C and resolution for at = 1.0 and a = 0
Chapter 3 Testing
Relative Amplitude o .
Figure 37. Relative amplitude versus C and resolution for at = 1.0 and
a = 0.25
Figure 38. Relative speed versus C and resolution for at = 1 .O and a = 0.25
Chapter 3 Testing
Relative Amplitude o.
Figure 39. Relative amplitude versus C and resolution for at = 1.5 and a = 0
Relative Speed 0.
Figure 40. Relative speed versus C and resolution for at = 1.5 and a = 0
Chapter 3 Testing
Relative Amplitude
Figure 41. Relative amplitude versus C and resolution for at = 1.5 and
a = 0.25
Figure 42. Relative speed versus C and resolution for at = 1.5 and a = 0.25
Chapter 3 Testing
4 Conclusions
The test cases demonstrate that the resulting model converges to the correct
heights and shock speeds with increasing resolution. Furthermore, general 2-D
cases of lateral transition in supercritical flow showed the model to compare
quite well in reproducing the oblique shock pattern.
The trigger mechanism, based upon energy variation, appears to detect the
jump quite well. The Petrov-Galerkin technique shown is an intuitive method
relying upon characteristic speeds and directions and produces a 2-D model
which is adequate to address hydraulic problems involving jumps and oblique
shocks.
Chapter 4 Discussion
References
Courant, R., and Friedrichs, K. 0 . (1948). Supersonic flow and shock waves,
Interscience Publishers, New York, 121-126.
Courant, R., Isaacson, E., and Rees, M. (1952). "On the solution of nonlinear
hyperbolic differential equations," Communication on Pure and Applied
Mathematics 5, 243-255.
Gabutti, B. (1983). "On two upwind finite different schemes for hyperbolic
equations in non-conservative form," Coinputers and Fluids 11(3), 207-230.
References
Hughes, T. J. R., and Brooks, A. N. (1982). "A theoretical framework for
Petrov-Galerkin methods with discontinuous weighting functions: Applica-
tions to the streamline-upwind procedures." Finite Elements in Fluids.
R. H. Gallagher, et al., ed., J. Wiley and Sons, London, 4, 47-65.
Von Neumann, J., and Richtmyer, R. D. (1950). "A method for the numerical
calculation of hydrodynamic shocks," Journal of Applied Physics 21, 232-
237.
References
Form Approved
REPORT DOCUMENTATION PAGE OMB NO. 0704-0188
I"
vailable from National Technical Information Service, 5285 Port Royal Road, Springfield, VA 22161.
ing up O(1) errors, but restricting the error to the neighborhood of the jump or shock. This technique is called
ction matrix. Furthermore, in order to restrict the shock capturing to the vicinity of the jump, a method of
detection is implemented which depends on the variation of mechanical energy within an element.
The veracity of the model is tested by comparison of the predicted jump speed and magnitude with
nalytic and flume results. A comparison is also made to a flume case of steady-state supercritical lateral